Hybrid BSS techniques for fetal ECG extraction using a semi-synthetic database

Noninvasive fetal ECG monitoring makes use of electrodes placed on the mother's abdomen with sophisticated algorithms to separate fetal and mother activity. In this task often Blind Source Separation is used as an intermediate step. To solve Blind Source Separation problem is generally employed PCA and ICA. COMBI and MULTI-COMBI algorithms offer novel schemes for combining PCA and ICA, enabling exploit the strengths of both techniques. In this work, the performance of the algorithms COMBI, MULTICOMBI, and traditional JADE algorithm are compared. We used a semi-synthetic database compound of 26 abdominal ECG records, where real 3-D ECG signals representing maternal and fetal cardiac components are mixed by volume conduction transfer matrices, in which, the fetal to mother ratio SIR and fetal to noise ratio SNR are controlled. As performance parameter, we used the signal to error relation SER, varying SIR between -5dB to -30dB, and SNR between 0dB to 30dB in all possible combinations. In all case, white noise or pink noise included, it is found that the COMBI and MULTICOMBI algorithms show better performance than the JADE algorithm.